Considering an enormous semantic gap problem between the low-level visual features and high-level semantic information of images, and the fact that the accuracy of content-based image classification and retrieval depends greatly on the description of low-level visual features, an image semantic classification approach is proposed based on Multiple-hyperplanes Support Vector Machines (MHSVMs). The multiple-hyperplane classifier, which is investigated from the complexity of optimization problem and the generalization performance, is the explicit extension of the optimal separating hyperplanes classifier. Experimental results show that the proposed approach is more accurate in image semantic classification than other ones, such as SVMs classifier using color and textural features.